opendatalab/MinerU ? reverse-engineered prompt

I need a Python tool that can process complex documents and get them ready for an AI to use. It should handle formats like PDFs, Word documents, and PowerPoint presentations. The main problem with these files is that their layout is complicated, with tables, columns, and headers that get messed up if you just copy the text.

This tool should be smart enough to properly analyze the document's structure, correctly identifying titles, lists, and especially tables, preserving their rows and columns. It also needs to have OCR to extract text from scanned documents or images. The goal is to convert these files into clean, well-structured markdown or JSON. This will make the content much easier to feed into a large language model for analysis or use in a RAG pipeline.